This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. 529 Chapter 16 CHAPTER 16 HTTP Headers for Optimal Performance Header composition is often neglected in the CGI world. Dynamic content is dynamic, after all, so why would anybody care about HTTP headers? Because pages are gener- ated dynamically, one might expect that pages without a Last-Modified header are fine, and that an If-Modified-Since header in the client’s request can be ignored. This laissez-faire attitude is a disadvantage when you’re trying to create a server that is entirely driven by dynamic components and the number of hits is significant. If the number of hits on your server is not significant and is never going to be, then it is safe to skip this chapter. But if keeping up with the number of requests is impor- tant, learning what cache-friendliness means and how to cooperate with caches to increase the performance of the site can provide significant benefits. If Squid or mod_proxy is used in httpd accelerator mode (as discussed in Chapter 12), it is cru- cial to learn how best to cooperate with it. In this chapter, when we refer to a section in the HTTP standard, we are using HTTP standard 1.1, which is documented in RFC 2616. The HTTP standard describes many headers. In this chapter, we discuss only the headers most relevant to caching. We divide them into three sets: date headers, content headers, and the special Vary header. Date-Related Headers The various headers related to when a document was created, when it was last modi- fied, and when it should be considered stale are discussed in the following sections. Date Header Section 14.18 of the HTTP standard deals with the circumstances under which we must or must not send a Date header. For almost everything a normal mod_perl user does, a Date header needs to be generated. But the mod_perl programmer doesn’t have to worry about this header, since the Apache server guarantees that it is always sent. ,ch16.24742 Page 529 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. 530 | Chapter 16: HTTP Headers for Optimal Performance In http_protocol.c, the Date header is set according to $r->request_time. A mod_perl script can read, but not change, $r->request_time. Last-Modified Header Section 14.29 of the HTTP standard covers the Last-Modified header, which is mostly used as a weak validator. Here is an excerpt from the HTTP specification: A validator that does not always change when the resource changes is a "weak validator." One can think of a strong validator as one that changes whenever the bits of an entity changes, while a weak value changes whenever the meaning of an entity changes. What this means is that we must decide for ourselves when a page has changed enough to warrant the Last-Modified header being updated. Suppose, for example that we have a page that contains text with a white background. If we change the background to light gray then clearly the page has changed, but if the text remains the same we would consider the semantics (meaning) of the page to be unchanged. On the other hand, if we changed the text, the semantics may well be changed. For some pages it is not quite so straightforward to decide whether the semantics have changed or not. This may be because each page comprises several components, or it might be because the page itself allows interaction that affects how it appears. In all cases, we must determine the moment in time when the semantics changed and use that moment for the Last-Modified header. Consider for example a page that provides a text-to-GIF renderer that takes as input a font to use, background and foreground colors, and a string to render. The images embedded in the resultant page are generated on the fly, but the structure of the page is constant. Should the page be considered unchanged so long as the underlying script is unchanged, or should the page be considered to have changed with each new request? Actually, a few more things are relevant: the semantics also change a little when we update one of the fonts that may be used or when we update the ImageMagick or equivalent image-generating program. All the factors that affect the output should be considered if we want to get it right. In the case of a page comprised of several components, we must check when the semantics of each component last changed. Then we pick the most recent of these times. Of course, the determination of the moment of change for each component may be easy or it may be subtle. mod_perl provides two convenient methods to deal with this header: update_mtime( ) and set_last_modified( ). These methods and several others are unavailable in the standard mod_perl environment but are silently imported when we use Apache:: File . Refer to the Apache::File manpage for more information. ,ch16.24742 Page 530 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. Date-Related Headers | 531 The update_mtime( ) function takes Unix’s time(2) (in Perl the equivalent is also the time( ) function) as its argument and sets Apache’s request structure finfo.st_mtime to this value. It does so only when the argument is greater than the previously stored finfo.st_mtime. The set_last_modified( ) function sets the outgoing Last-Modified header to the string that corresponds to the stored finfo.st_mtime. When passing a Unix time(2) to set_last_modified( ), mod_perl calls update_mtime( ) with this argument first. The following code is an example of setting the Last-Modified header by retrieving the last-modified time from a Revision Control System (RCS)–style of date tag. use Apache::File; use Date::Parse; $Mtime ||= Date::Parse::str2time( substr q$Date: 2003/05/09 21:34:23 $, 6); $r->set_last_modified($Mtime); Normally we would use the Apache::Util::parsedate function, but since it doesn’t parse the RCS format, we have used the Date::Parse module instead. Expires and Cache-Control Headers Section 14.21 of the HTTP standard deals with the Expires header. The purpose of the Expires header is to determine a point in time after which the document should be considered out of date (stale). Don’t confuse this with the very different meaning of the Last-Modified header. The Expires header is useful to avoid unnecessary vali- dation from now until the document expires, and it helps the recipients to clean up their stored documents. Here’s an excerpt from the HTTP standard: The presence of an Expires field does not imply that the original resource will change or cease to exist at, before, or after that time. Think carefully before setting up a time when a resource should be regarded as stale. Most of the time we can determine an expected lifetime from “now” (that is, the time of the request). We do not recommend hardcoding the expiration date, because when we forget that we did it, and the date arrives, we will serve already expired doc- uments that cannot be cached. If a resource really will never expire, make sure to fol- low the advice given by the HTTP specification: To mark a response as "never expires," an origin server sends an Expires date approximately one year from the time the response is sent. HTTP/1.1 servers SHOULD NOT send Expires dates more than one year in the future. For example, to expire a document half a year from now, use the following code: $r->header_out('Expires', HTTP::Date::time2str(time + 180*24*60*60)); or: $r->header_out('Expires', Apache::Util::ht_time(time + 180*24*60*60)); ,ch16.24742 Page 531 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. 532 | Chapter 16: HTTP Headers for Optimal Performance The latter method should be faster, but it’s available only under mod_perl. A very handy alternative to this computation is available in the HTTP/1.1 cache-con- trol mechanism. Instead of setting the Expires header, we can specify a delta value in a Cache-Control header. For example: $r->header_out('Cache-Control', "max-age=" . 180*24*60*60); This is much more processor-economical than the previous example because Perl computes the value only once, at compile time, and optimizes it into a constant. As this alternative is available only in HTTP/1.1 and old cache servers may not under- stand this header, it may be advisable to send both headers. In this case the Cache- Control header takes precedence, so the Expires header is ignored by HTTP/1.1-com- pliant clients. Or we could use an if else clause: if ($r->protocol =~ /(\d\.\d)/ && $1 >= 1.1) { $r->header_out('Cache-Control', "max-age=" . 180*24*60*60); } else { $r->header_out('Expires', HTTP::Date::time2str(time + 180*24*60*60)); } Again, use the Apache::Util::ht_time( ) alternative instead of HTTP::Date:: time2str( ) if possible. If the Apache server is restarted regularly (e.g., for log rotation), it might be benefi- cial to save the Expires header in a global variable to save the runtime computation overhead. To avoid caching altogether, call: $r->no_cache(1); which sets the headers: Pragma: no-cache Cache-control: no-cache This should work in most browsers. Don’t set Expires with $r->header_out if you use $r->no_cache, because header_out( ) takes precedence. The problem that remains is that there are broken browsers that ignore Expires headers. Content Headers The following sections describe the HTTP headers that specify the type and length of the content, and the version of the content being sent. Note that in this section we often use the term message. This term is used to describe the data that comprises the HTTP headers along with their associated content; the content is the actual page, image, file, etc. ,ch16.24742 Page 532 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. Content Headers | 533 Content-Type Header Most CGI programmers are familiar with Content-Type. Sections 3.7, 7.2.1, and 14.17 of the HTTP specification cover the details. mod_perl has a content_type( ) method to deal with this header: $r->content_type("image/png"); Content-Type should be included in every set of headers, according to the standard, and Apache will generate one if your code doesn’t. It will be whatever is specified in the relevant DefaultType configuration directive, or text/plain if none is active. Content-Length Header According to section 14.13 of the HTTP specification, the Content-Length header is the number of octets (8-bit bytes) in the body of a message. If the length can be determined prior to sending, it can be very useful to include it. The most important reason is that KeepAlive requests (when the same connection is used to fetch more than one object from the web server) work only with responses that contain a Content-Length header. In mod_perl we can write: $r->header_out('Content-Length', $length); When using Apache::File, the additional set_content_length( ) method, which is slightly more efficient than the above, becomes available to the Apache class. In this case we can write: $r->set_content_length($length); The Content-Length header can have a significant impact on caches by invalidating cache entries, as the following extract from the specification explains: The response to a HEAD request MAY be cacheable in the sense that the information contained in the response MAY be used to update a previously cached entity from that resource. If the new field values indicate that the cached entity differs from the current entity (as would be indicated by a change in Content-Length, Content-MD5, ETag or Last-Modified), then the cache MUST treat the cache entry as stale. It is important not to send an erroneous Content-Length header in a response to either a GET or a HEAD request. Entity Tags An entity tag (ETag) is a validator that can be used instead of, or in addition to, the Last-Modified header; it is a quoted string that can be used to identify different ver- sions of a particular resource. An entity tag can be added to the response headers like this: $r->header_out("ETag","\"$VERSION\""); ,ch16.24742 Page 533 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. 534 | Chapter 16: HTTP Headers for Optimal Performance mod_perl offers the $r->set_etag( ) method if we have use( )ed Apache::File. How- ever, we strongly recommend that you don’t use the set_etag( ) method! set_etag( ) is meant to be used in conjunction with a static request for a file on disk that has been stat( )ed in the course of the current request. It is inappropriate and danger- ous to use it for dynamic content. By sending an entity tag we are promising the recipient that we will not send the same ETag for the same resource again unless the content is “equal” to what we are sending now. The pros and cons of using entity tags are discussed in section 13.3 of the HTTP spec- ification. For mod_perl programmers, that discussion can be summed up as follows. There are strong and weak validators. Strong validators change whenever a single bit changes in the response; i.e., when anything changes, even if the meaning is unchanged. Weak validators change only when the meaning of the response changes. Strong validators are needed for caches to allow for sub-range requests. Weak valida- tors allow more efficient caching of equivalent objects. Algorithms such as MD5 or SHA are good strong validators, but what is usually required when we want to take advantage of caching is a good weak validator. A Last-Modified time, when used as a validator in a request, can be strong or weak, depending on a couple of rules described in section 13.3.3 of the HTTP standard. This is mostly relevant for range requests, as this quote from section 14.27 explains: If the client has no entity tag for an entity, but does have a Last-Modified date, it MAY use that date in an If-Range header. But it is not limited to range requests. As section 13.3.1 states, the value of the Last- Modified header can also be used as a cache validator. The fact that a Last-Modified date may be used as a strong validator can be pretty disturbing if we are in fact changing our output slightly without changing its seman- tics. To prevent this kind of misunderstanding between us and the cache servers in the response chain, we can send a weak validator in an ETag header. This is possible because the specification states: If a client wishes to perform a sub-range retrieval on a value for which it has only a Last-Modified time and no opaque validator, it MAY do this only if the Last- Modified time is strong in the sense described here. In other words, by sending an ETag that is marked as weak, we prevent the cache server from using the Last-Modified header as a strong validator. An ETag value is marked as a weak validator by prepending the string W/ to the quoted string; otherwise, it is strong. In Perl this would mean something like this: $r->header_out('ETag',"W/\"$VERSION\""); ,ch16.24742 Page 534 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. Content Headers | 535 Consider carefully which string is chosen to act as a validator. We are on our own with this decision: only the service author knows the semantics of a resource well enough to select an appropriate cache validation mechanism, and the specification of any validator comparison function more complex than byte-equality would open up a can of worms. Thus, comparisons of any other headers (except Last-Modified, for compatibility with HTTP/1.0) are never used for purposes of validating a cache entry. If we are composing a message from multiple components, it may be necessary to combine some kind of version information for all these components into a single string. If we are producing relatively large documents, or content that does not change fre- quently, then a strong entity tag will probably be preferred, since this will give caches a chance to transfer the document in chunks. HTTP Range Requests It is possible in web clients to interrupt the connection before the data transfer has fin- ished. As a result, the client may have partial documents or images loaded into its memory. If the page is reentered later, it is useful to be able to request the server to return just the missing portion of the document, instead of retransferring the entire file. There are also a number of web applications that benefit from being able to request the server to give a byte range of a document. As an example, a PDF viewer would need to be able to access individual pages by byte range—the table that defines those ranges is located at the end of the PDF file. In practice, most of the data on the Web is represented as a byte stream and can be addressed with a byte range to retrieve a desired portion of it. For such an exchange to happen, the server needs to let the client know that it can sup- port byte ranges, which it does by sending the Accept-Ranges header: Accept-Ranges: bytes The server will send this header only for documents for which it will be able to satisfy the byte-range request—e.g., for PDF documents or images that are only partially cached and can be partially reloaded if the user interrupts the page load. The client requests a byte range using the Range header: Range: bytes=0-500,5000- Because of the architecture of the byte-range request and response, the client is not lim- ited to attempting to use byte ranges only when this header is present. If a server does not support the Range header, it will simply ignore it and send the entire document as a response. ,ch16.24742 Page 535 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. 536 | Chapter 16: HTTP Headers for Optimal Performance Content Negotiation Content negotiation is a wonderful feature that was introduced with HTTP/1.1. Unfortunately it is not yet widely supported. Probably the most popular usage sce- nario for content negotiation is language negotiation for multilingual sites. Users specify in their browsers’ preferences the languages they can read and order them according to their ability. When the browser sends a request to the server, among the headers it sends it also includes an Accept-Language header. The server uses the Accept-Language header to determine which of the available representations of the document best fits the user’s preferences. But content negotiation is not limited to language. Quoting the specification: HTTP/1.1 includes the following request-header fields for enabling server-driven negotiation through description of user agent capabilities and user preferences: Accept (section 14.1), Accept-Charset (section 14.2), Accept-Encoding (section 14.3), Accept-Language (section 14.4), and User-Agent (section 14.43). However, an origin server is not limited to these dimensions and MAY vary the response based on any aspect of the request, including information outside the request-header fields or within extension header fields not defined by this specification. The Vary Header To signal to the recipient that content negotiation has been used to determine the best available representation for a given request, the server must include a Vary header. This tells the recipient which request headers have been used to determine the representation that is used. So an answer may be generated like this: $r->header_out('Vary', join ", ", qw(accept accept-language accept-encoding user-agent)); The header of a very cool page may greet the user with something like this: Hallo Harri, Dein NutScrape versteht zwar PNG aber leider kein GZIP. However, this header has the side effect of being expensive for a caching proxy. As of this writing, Squid (Version 2.3.STABLE4) does not cache resources that come with a Vary header at all. So without a clever workaround, the Squid accelerator is of no use for these documents. HTTP Requests Section 13.11 of the specification states that the only two cacheable methods are GET and HEAD. Responses to POST requests are not cacheable, as you’ll see in a moment. ,ch16.24742 Page 536 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. HTTP Requests | 537 GET Requests Most mod_perl programs are written to service GET requests. The server passes the request to the mod_perl code, which composes and sends back the headers and the content body. But there is a certain situation that needs a workaround to achieve better cacheabil- ity. We need to deal with the "?" in the relative path part of the requested URI. Sec- tion 13.9 specifies that: caches MUST NOT treat responses to such URIs as fresh unless the server provides an explicit expiration time. This specifically means that responses from HTTP/1.0 servers for such URIs SHOULD NOT be taken from a cache. Although it is tempting to imagine that if we are using HTTP/1.1 and send an explicit expiration time we are safe, the reality is unfortunately somewhat different. It has been common for quite a long time to misconfigure cache servers so that they treat all GET requests containing a question mark as uncacheable. People even used to mark anything that contained the string “cgi-bin” as uncacheable. To work around this bug in HEAD requests, we have stopped calling CGI directories cgi-bin and we have written the following handler, which lets us work with CGI-like query strings without rewriting the software (e.g., Apache::Request and CGI.pm) that deals with them: sub handler { my $r = shift; my $uri = $r->uri; if ( my($u1,$u2) = $uri =~ / ^ ([^?]+?) ; ([^?]*) $ /x ) { $r->uri($u1); $r->args($u2); } elsif ( my ($u1,$u2) = $uri =~ m/^(.*?)%3[Bb](.*)$/ ) { # protect against old proxies that escape volens nolens # (see HTTP standard section 5.1.2) $r->uri($u1); $u2 =~ s/%3[Bb]/;/g; $u2 =~ s/%26/;/g; # & $u2 =~ s/%3[Dd]/=/g; $r->args($u2); } DECLINED; } This handler must be installed as a PerlPostReadRequestHandler. The handler takes any request that contains one or more semicolons but no question mark and changes it so that the first semicolon is interpreted as a question mark and everything after that as the query string. So now we can replace the request: http://example.com/query?BGCOLOR=blue;FGCOLOR=red ,ch16.24742 Page 537 Thursday, November 18, 2004 12:43 PM This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc. All rights reserved. 538 | Chapter 16: HTTP Headers for Optimal Performance with: http://example.com/query;BGCOLOR=blue;FGCOLOR=red This allows the coexistence of queries from ordinary forms that are being processed by a browser alongside predefined requests for the same resource. It has one minor bug: Apache doesn’t allow percent-escaped slashes in such a query string. So instead of: http://example.com/query;BGCOLOR=blue;FGCOLOR=red;FONT=%2Ffont%2Fpath we must use: http://example.com/query;BGCOLOR=blue;FGCOLOR=red;FONT=/font/path To unescape the escaped characters, use the following code: s/%([0-9A-Fa-f]{2})/chr hex $1/ge; Conditional GET Requests A rather challenging request that may be received is the conditional GET, which typi- cally means a request with an If-Modified-Since header. The HTTP specification has this to say: The semantics of the GET method change to a "conditional GET" if the request message includes an If-Modified-Since, If-Unmodified-Since, If-Match, If-None-Match, or If- Range header field. A conditional GET method requests that the entity be transferred only under the circumstances described by the conditional header field(s). The conditional GET method is intended to reduce unnecessary network usage by allowing cached entities to be refreshed without requiring multiple requests or transferring data already held by the client. So how can we reduce the unnecessary network usage in such a case? mod_perl makes it easy by providing access to Apache’s meets_conditions( ) function (which lives in Apache::File). The Last-Modified (and possibly ETag) headers must be set up before calling this method. If the return value of this method is anything other than OK, then this value is the one that should be returned from the handler when we have finished. Apache handles the rest for us. For example: if ((my $result = $r->meets_conditions) != OK) { return $result; } #else go and send the response body If we have a Squid accelerator running, it will often handle the conditionals for us, and we can enjoy its extremely fast responses for such requests by reading the access.log file. Just grep for TCP_IMS_HIT/304. However, there are circumstances under which Squid may not be allowed to use its cache. That is why the origin server (which is the server we are programming) needs to handle conditional GETs as well, even if a Squid accelerator is running. ,ch16.24742 Page 538 Thursday, November 18, 2004 12:43 PM [...]... browser or proxy, of an individual page wherein the page in question has data that may be of a sensitive nature (as in a “form page for submittal”) and the creator of the page wants to make sure that the page does not get submitted twice 540 | Chapter 16: HTTP Headers for Optimal Performance This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc All rights reserved...,ch16.24742 Page 539 Thursday, November 18, 2004 12:43 PM HEAD Requests Among the headers described thus far, the date-related ones (Date, Last-Modified, and Expires/Cache-Control) are usually easy to produce and thus should be computed for HEAD requests just the same as for GET requests The Content-Type and Content-Length headers should be exactly the same as would be supplied to the corresponding GET... reserved 539 ,ch16.24742 Page 540 Thursday, November 18, 2004 12:43 PM lots of requests to retrieve the same content before it changes again, and if it is much cheaper to test whether the content needs refreshing than it is to refresh it In this situation, a PerlFixupHandler can be installed for the relevant location This handler must test whether the content is up to date or not, returning DECLINED so that... content has expired, the handler should regenerate the content into the file, update the $r->finfo status and still return DECLINED, which will force Apache to serve the now updated file Updating $r->finfo can be achieved by calling: $r->filename($file); # force update of the finfo structure even if this seems redundant because the filename is the same as $file This is important because otherwise Apache... visibility issues that apply to GETs, so they may not be suitable in every case Avoiding Dealing with Headers There is another approach to dynamic content that is possible with mod_perl This approach is appropriate if the content changes relatively infrequently, if we expect Avoiding Dealing with Headers | This is the Title of the Book, eMatter Edition Copyright © 2004 O’Reilly & Associates, Inc All... itself, but under some circumstances it may not be allowed to do so POST Requests The response to a POST request is not cacheable, due to an underspecification in the HTTP standards Section 13.4 does not forbid caching of responses to POST requests, but no other part of the HTTP standard explains how the caching of POST requests could be implemented, so we are in a vacuum No existing caching servers implement... caching of POST requests (although some browsers with more aggressive caching implement their own caching of POST requests) However, this may change if someone does the groundwork of defining the semantics for cache operations on POST requests Note that if a Squid accelerator is being used, you should be aware that it accelerates outgoing traffic but does not bundle incoming traffic Squid is of no benefit... the specification that requires them to be sent What is important is that the response to a HEAD request must not contain a messagebody The code in a mod_perl handler might look like this: # compute the headers that are easy to compute # currently equivalent to $r->method eq "HEAD" if ( $r->header_only ) { $r->send_http_header; return OK; } If a Squid accelerator is being used, it will be able to handle . reserved. 530 | Chapter 16: HTTP Headers for Optimal Performance In http_protocol.c, the Date header is set according to $r->request_time. A mod_ perl script can. reserved. 532 | Chapter 16: HTTP Headers for Optimal Performance The latter method should be faster, but it’s available only under mod_ perl. A very handy alternative